But appropriately accounted for, I think it is.
The relevant metric would be, "what do I have to increase in order to get more [interesting, new] theorems proved?" Is it as easy has having mathematicians work more hours? Having more people start working on them?
Intuitively, it is not so easy -- each insights need exponentially more work as time progresses. An exponentially-small fraction of humans is capable of producing novel results, and so on.
This, I think, is what Aaronson is getting at: if P = NP, if proving is as easy as verifying, then coming up with hard mathematical insights should be as easy as following a cookbook, or any other routine, mundane mental task.
But that doesn't seem to describe the world we live in, where getting mathematical insights has rapidly diminishing returns per unit resource invested.
What we haven't automated is mathematical creativity, which is a completely different class of problem.
Creativity is an anthropomorphic, abstract concept, a way of describing human ways to find solutions. What really matters is the proofs.
I have a problem with this. Just because the thought is unpleasant does not make it true or false. Besides, perhaps this is the very difference between automation and intelligence, perhaps the point at which you can ask this of a computer is the point at which you should no longer consider it a computer.